Predicting Carcinogenicity of Polycyclic Aromatic Hydrocarbons from Back-Propagation Neural Network

Models of relationships between structure and carcinogenicity of polycyclic aromatic hydrocarbons were constructed by means of a multilayer neural network using the back-propagation algorithm. The molecular descriptors used were derived from graph theory. The neural network (NN) was used to classify the compounds studied into two categories, namely inactive or active. To evaluate the predictive power of an NN model, the cross-validation procedure was used. The total prediction accuracy of 86% (90% of the actives correctly identified) provided an evidence of the usefulness of the present neural algorithm. © 1994, American Chemical Society. All rights reserved.